[Research on modelling vestibular rehabilitation decision based on machine learning]

Area of Science:

• Neuroscience
• Medical Informatics
• Machine Learning

Background:

• Vestibular rehabilitation is crucial for managing dizziness and balance disorders.
• Current treatment selection can be complex, relying on subjective assessments.
• Objective data analysis may improve rehabilitation program efficacy.

Purpose of the Study:

• To compare the effectiveness of Support Vector Machine (SVM) and Artificial Neural Network (ANN) models in selecting vestibular rehabilitation strategies.
• To assess the accuracy of machine learning algorithms in predicting optimal treatment plans.

Main Methods:

• Utilized Sensory Organization Test (SOT) scores (somatosensory, visual, vestibular) and Dizziness Handicap Inventory (DHI) sub-scores (physical, emotional, functional) as input features.
• Trained both SVM and ANN models using a simulated database and clinical data.
• Evaluated model performance based on rehabilitation program output accuracy.

Main Results:

• The SVM model achieved an accuracy rate of 83.4%, significantly outperforming the ANN model's 52.3%.
• Model errors were primarily attributed to overlapping score data intervals across diagnostic schemes, leading to misclassification of boundary cases.
• SVM demonstrated superior performance in distinguishing between rehabilitation needs.

Conclusions:

• Support Vector Machine (SVM) provides a more accurate approach than Artificial Neural Network (ANN) for computer-assisted vestibular rehabilitation decision-making.
• Machine learning integration holds significant promise for advancing clinical informatization and enhancing the quality of vestibular disorder treatment.
• Further research is warranted to refine algorithms and address data overlap challenges in clinical practice.